WISE: A Lightweight, Weakly-Supervised Model for Onboard Fire Smoke Detection and Localization

Sha Lu; Yu Sun; Liang Zhao; Jixue Liu; Lin Liu; Jiuyong Li; A. K. Qin; Alejandro Mousist

cs.CVSep 16, 2026

WISE: A Lightweight, Weakly-Supervised Model for Onboard Fire Smoke Detection and Localization

Authors: Sha Lu, Yu Sun, Liang Zhao, Jixue Liu, Lin Liu, Jiuyong Li, A. K. Qin, Alejandro Mousist,

Organizations: Adelaide University, Adelaide, SA 5000, Australia · Swinburne University of Technology, Hawthorn, VIC 3122, Australia · Thales Alenia Space, Cannes, 06150, France

Abstract

Wildfire smoke detection from satellite imagery is critical for early warning and rapid response. For onboard satellite deployment, detection systems must operate under strict memory and latency constraints while providing spatially informative outputs for downstream decision-making. Existing tile-level classification methods are computationally efficient but lack spatial localization, whereas pixel-level segmentation approaches provide detailed masks yet are typically too computationally demanding for real-time onboard execution. To address this gap, we propose WISE (Weakly-supervised Inference-efficient Smoke Extraction), a deployment-oriented framework for onboard fire smoke detection and localization. WISE leverages only tile-level annotations through a teacher-student distillation strategy, where an offline teacher provides soft spatial supervision to a lightweight WISE-Student optimized for efficient onboard inference. The student jointly predicts tile-level smoke presence and smoke probability maps within a single forward pass, enabling spatially informative detection under strict computational constraints. WISE was evaluated through in-orbit execution aboard the ISS-mounted IMAGIN-e payload. Three model variants achieve average inference times of 0.10 s, 0.14 s, and 0.26 s per tile, indicating near-real-time per-tile inference within onboard resource limits. Ground-based experiments on Landsat 5 and Landsat 8 imagery further indicate effective detection and spatially informative localization. The best-performing variant achieves a mean tile-level F1 score of 0.964 and a mean pixel-level F1 score of 0.750 across 10 runs, while containing only 0.12M parameters and requiring approximately 3 GFLOPs. Together, these results indicate that WISE is a practical candidate for low-latency wildfire smoke monitoring from space under onboard resource constraints.

Figures & tables

Figure 1: Overall WISE pipeline. A teacher model generates Grad-CAM attention maps, which guide training of the lightweight WISE-Student for joint smoke detection and localization during onboard inference.

Figure 2: Multi-layer Grad-CAM activation maps produced by the teacher model trained with tile-level supervision. Grad-CAM maps from Layers 2–4 and their fused result illustrate the emergence of spatial attention from classification-only training.

Figure 3: Architecture of the WISE-Student model. The network jointly produces tile-level smoke prediction and a high-resolution smoke probability map in a single forward pass, with dashed components used only during training.

Model	ResNet-50	WISE-T	WISE-S	WISE-M	WISE-L
Parameters (M)	24.57	0.008	0.030	0.066	0.116
FLOPs (G)	5.566	0.228	0.810	1.748	3.042

Table 1: Comparison of parameter count and FLOPs for the ResNet-50 teacher and WISE model variants.

Metric	WISE-T	WISE-S	WISE-M	WISE-L	ResNet-50
Accuracy	$0.840\pm 0.069$ (0.910)	$0.959\pm 0.014$ (0.984)	$0.975\pm 0.009$ (0.992)	$\mathbf{0.976\pm 0.006}$ (0.986)	0.986
F1 Score	$0.767\pm 0.081$ (0.865)	$0.939\pm 0.021$ (0.976)	$0.962\pm 0.014$ (0.988)	$\mathbf{0.964\pm 0.009}$ (0.980)	0.979
Precision	$0.776\pm 0.105$ (0.869)	$0.939\pm 0.025$ (0.974)	$0.966\pm 0.014$ (0.984)	$\mathbf{0.972\pm 0.015}$ (0.992)	1.000
Recall	$0.776\pm 0.115$ (0.951)	$0.938\pm 0.029$ (0.992)	$\mathbf{0.959\pm 0.020}$ (0.992)	$0.957\pm 0.017$ (0.976)	0.959
FNR	$0.224\pm 0.115$ (0.049)	$0.062\pm 0.029$ (0.008)	$\mathbf{0.041\pm 0.020}$ (0.008)	$0.043\pm 0.017$ (0.024)	0.041

Table 2: Classification performance of WISE variants and the teacher ResNet-50 model. Each cell reports the mean

\pm

standard deviation across 10 runs, and the best result achieved in any run is shown in parentheses. For WISE variants, the best average result for each metric is highlighted in bold.

Model	Params (M)	FLOPs (G)	Accuracy	F1	FNR
CloudScout	1.263	3.533	$\mathbf{0.985}$	$\mathbf{0.978}$	$\mathbf{0.023}$
MobileNetV2	2.882	0.450	$0.964$	$0.944$	$0.074$
PhiFire	26.094	0.913	$0.926$	$0.888$	$0.120$
VIB-SD	0.550	0.651	$0.983$	$0.975$	$0.031$
WISE-M (Ours)	0.066	1.748	$0.975$	$0.962$	$0.041$

Table 3: Tile-level classification performance and efficiency comparison between WISE-M and lightweight baseline models. Classification results are reported as mean values across 10 runs. The best performance among the baseline models is highlighted in bold.

Metric	WISE-T	WISE-S	WISE-M	WISE-L	ResNet-50
Accuracy	$0.743\pm 0.10$ (0.875)	$0.828\pm 0.03$ (0.878)	$0.851\pm 0.03$ (0.894)	$\mathbf{0.866}\pm 0.04$ ( 0.915 )	$0.778$
F1 Score	$0.601\pm 0.07$ (0.707)	$0.697\pm 0.03$ (0.754)	$0.730\pm 0.03$ (0.778)	$\mathbf{0.750}\pm 0.04$ ( 0.806 )	$0.605$
IoU	$0.461\pm 0.07$ (0.593)	$0.563\pm 0.03$ (0.620)	$0.600\pm 0.04$ (0.655)	$\mathbf{0.627}\pm 0.04$ ( 0.692 )	$0.463$
Precision	$0.569\pm 0.09$ (0.707)	$0.676\pm 0.04$ (0.739)	$0.714\pm 0.04$ (0.767)	$\mathbf{0.733}\pm 0.04$ ( 0.789 )	$0.514$
Recall	$0.759\pm 0.08$ (0.815)	$0.777\pm 0.04$ (0.828)	$0.795\pm 0.04$ (0.838)	$\mathbf{0.808}\pm 0.03$ ( 0.852 )	$0.856$

Table 4: Pixel-level performance of WISE variants and the ResNet Grad-CAM baseline. Each cell reports the mean

\pm

standard deviation across 10 runs, and the best result achieved in any run is shown in parentheses. The best result for each metric is highlighted in bold.

Metric	VTrUNet	SUNet	FrizCNN	WISE-L (Best Run)
F1	0.826	0.768	0.691	0.806
Precision	0.833	0.739	0.666	0.789
Recall	0.889	0.830	0.770	0.852

Table 5: Pixel-level performance comparison between WISE-L and supervised segmentation models. Best results for each model are reported; the WISE-L value is the best result from a single run in Table 4 , whose 10-run mean F1 is 0.750.

Figure 4: Qualitative results of WISE-L on ten representative examples (A–J). Each example shows the RGB input, the teacher Grad-CAM attention map, and the predicted smoke probability map, with red polygons indicating manually annotated smoke regions.

Model	Grad-CAM Supervision	Tile-Level		Pixel-Level
	Grad-CAM Supervision	F1	FNR	F1
WISE-T	With	0.767	0.224	0.601
WISE-T	Without	0.860	0.143	0.406
WISE-S	With	0.939	0.062	0.697
WISE-S	Without	0.910	0.107	0.404
WISE-M	With	0.962	0.041	0.730

Table 6: Impact of Grad-CAM supervision across WISE variants. Results are averaged over 10 runs. Pixel-level evaluation is reported only in terms of F1, as FNR is not commonly used in smoke detection due to the inherent difficulty of exhaustively labeling all smoke pixels.

Model	Batch Size	Overlap	#Tiles	Inference Time (s/tile)	Memory (Peak, MB)	Model Size (MB)
WISE-S	4	50	400	0.10	903	0.14
WISE-M	8	25	187	0.14	767	0.28
WISE-L	4	0	117	0.26	582	0.47

Table 7: Runtime performance of WISE models on the IMAGIN-e payload.

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Block	Operation	Input Shape	Output Shape
Input	-	$256\times 256\times 6$	$256\times 256\times 6$
Stem	Depthwise Separable Conv ( $k=3$ )	$256\times 256\times 6$	$256\times 256\times 24$
Shallow Encoder Block	Depthwise Separable Conv ( $k=3$ ) + Channel Attention	$256\times 256\times 24$	$256\times 256\times 48$
Downsampling Layer	Conv2D ( $k=3$ , stride=2) + GN + ReLU	$256\times 256\times 48$	$128\times 128\times 48$
Deep Encoder Block	Depthwise Separable Conv ( $k=3$ ) + Channel Attention	$128\times 128\times 48$	$128\times 128\times 96$
Lite-ASPP Module	$1\times 1$ Conv + 2 $\times$ Depthwise $3\times 3$ (dilated) + Projection	$128\times 128\times 96$	$128\times 128\times 48$

Appendix

Table A.1: Layer-wise configuration of the WISE model with width multiplier

\alpha=0.75

. The model uses a 6-channel

256\times 256

input and outputs both a tile-level class prediction and a smoke probability map.

Wildfire Detection

Wildfire detection uses satellite and aerial data with machine learning to identify active fires and smoke early enough to support response.

See all 20 papers

On-Orbit Real-Time Wildfire Detection Under On-Board Constraints
May 7, 2026Matthias Rötzer, Veronika Pörtge, Martin Ickerott +6Efficient Neural Network Inference Wildfire Detection
OroraTech GmbH, St.-Martin-Str. 112, 81669 Munich, Germany · Ludwig-Maximilians-Universität München, Munich, Germany
Construction and Evaluation of Machine Learning Models for Near-Real-Time Fire Detection from MTG FCI Imagery
Oct 4, 2026Asaf Vanunu, Boaz Nadler, Arnon KarnieliWildfire Detection Remote Sensing
Uncertainty-Aware Wildfire Smoke Density Classification from Satellite Imagery via CBAM-Augmented EfficientNet with Evidential Deep Learning
May 15, 2026Ranjith ChodavarapuRemote Sensing Image Understanding Evidential Deep Learning
Kent State University

May 7, 2026J/K move · Enter open · S save

On-Orbit Real-Time Wildfire Detection Under On-Board Constraints

Matthias Rötzer, Veronika Pörtge, Martin Ickerott, Jayendra Praveen Kumar Chorapalli+5

OroraTech GmbH, St.-Martin-Str. 112, 81669 Munich, Germany · Ludwig-Maximilians-Universität München, Munich, Germany

Efficient Neural Network Inference Wildfire Detection Infrared Small Target Detection Remote Sensing On-Device Inference Masked Image Modeling

Open paper AskPDF arXiv

We present a deployed system for on-orbit wildfire detection aboard a nine-satellite commercial thermal infrared constellation, operating under demanding joint constraints: sub-megabyte model footprint, sub-150 ms per-batch TensorRT FP16 inference on an NVIDIA Jetson Xavier NX, and an end-to-end alert pipeline targeting under 10 minutes from satellite overpass to fire event communication. The system operates on uncalibrated mid-wave infrared (MWIR) single-band imagery at 200 m ground sampling distance, where fires frequently appear as sub-pixel or single-pixel thermal anomalies under extreme class imbalance -- challenges not addressed by the contextual thermal-thresholding pipelines (MODIS, VIIRS) that currently dominate operational fire monitoring. We present an empirical study of lightweight dense representation learning for this regime using a proprietary nine-satellite MWIR dataset. We compare dense masked autoencoding (DenseMAE) and a hybrid DenseMAE+EMA (exponential moving average) distillation variant, and evaluate representations via linear probing and full-distribution pixel-level average precision (AP) under extreme class imbalance. DenseMAE pretraining enables compact downstream models on the latency-accuracy Pareto frontier: our fastest SSL-pretrained model achieves 0.640 test AP and 0.69 event-level Fire-F1 with 65.34 ms latency per batch and a 0.52 MB engine, without pruning or compression. The best configuration reaches 0.699 AP and 0.744 Fire-F1 below 1 MB, outperforming a supervised baseline (0.650 AP) under comparable constraints.

Abstract

Wildfire Detection

Explore similar work

On-Orbit Real-Time Wildfire Detection Under On-Board Constraints

Construction and Evaluation of Machine Learning Models for Near-Real-Time Fire Detection from MTG FCI Imagery

Uncertainty-Aware Wildfire Smoke Density Classification from Satellite Imagery via CBAM-Augmented EfficientNet with Evidential Deep Learning